If the test is run manually the output "error for solver ..."
is slightly misleading. In fact it's just the norm of the
achieved algebraic error. Hence "defect norm" is also
correct but less disturbing.